The Research Of Face Recognition Algorithm Based On Low Rank Recovery And Deep Neural Network

Face recognition is a hot topic in pattern recognition.Face recognition is widely used in authentication,detection and investigation of safety requirements of high places because of its convenience.Face recognition algorithm developed from the initial simple conditions of the identification to the identification of multiple factors under complex conditions.Face recognition in multiple factors in light,face rotation,face occlusion,noise pollution,skin color and race factors are the factors to be considered in the face recognition.Face recognition under complex condition is still a difficult problem in the field of face recognition.In this paper,the following issues are studied,and the solution is given.(1)The problem that wavelet threshold denoising can introduce quantization noise and improper selection threshold will damage the edge information of the image.In this paper,we proposed one algorithm based on wavelet threshold denoising and low rank matrix recovery.The experimental results show that proposed algorithm has better denoising effect than single wavelet thresholding to denoising algorithm,and it also improved the performance of the algorithm.(2)We aimed at the problem of excessive exposure,shadow and noise interference when the face images were collected.In this paper,we applied the low rank matrix recovery algorithm to the face image preprocessing.Experimental results showed that it improved the effects of face image effection of exposure,shadows,and interference.It improved the quality of the face image,laid the foundation for the subsequent extraction of high quality facial feature.(3)For the problem of extracting the poor robustness of facial features in the light of the linear dimension reduction method and the shallow layer neural network,in this paper,we combined the low rank matrix recovery algorithm with the depth neural network algorithm to solve this problem.We did experiments in YALE,ORL,AR face database,by setting the number of different network nodes and the number of iterations of the network,selecting the number of different training samples for the experiment.Experimental results on the face database show that the proposed algorithm has higher recognition rate and better stability than the linear dimension reduction and the shallow layer neural network algorithm.